1 /* SPDX-License-Identifier: GPL-2.0 */
2 #include <linux/linkage.h>
4 #include <asm/asm-offsets.h>
5 #include <asm/bitsperlong.h>
6 #include <asm/kvm_vcpu_regs.h>
7 #include <asm/nospec-branch.h>
8 #include <asm/percpu.h>
9 #include <asm/segment.h>
10 #include "run_flags.h"
12 #define WORD_SIZE (BITS_PER_LONG / 8)
14 #define VCPU_RAX __VCPU_REGS_RAX * WORD_SIZE
15 #define VCPU_RCX __VCPU_REGS_RCX * WORD_SIZE
16 #define VCPU_RDX __VCPU_REGS_RDX * WORD_SIZE
17 #define VCPU_RBX __VCPU_REGS_RBX * WORD_SIZE
18 /* Intentionally omit RSP as it's context switched by hardware */
19 #define VCPU_RBP __VCPU_REGS_RBP * WORD_SIZE
20 #define VCPU_RSI __VCPU_REGS_RSI * WORD_SIZE
21 #define VCPU_RDI __VCPU_REGS_RDI * WORD_SIZE
24 #define VCPU_R8 __VCPU_REGS_R8 * WORD_SIZE
25 #define VCPU_R9 __VCPU_REGS_R9 * WORD_SIZE
26 #define VCPU_R10 __VCPU_REGS_R10 * WORD_SIZE
27 #define VCPU_R11 __VCPU_REGS_R11 * WORD_SIZE
28 #define VCPU_R12 __VCPU_REGS_R12 * WORD_SIZE
29 #define VCPU_R13 __VCPU_REGS_R13 * WORD_SIZE
30 #define VCPU_R14 __VCPU_REGS_R14 * WORD_SIZE
31 #define VCPU_R15 __VCPU_REGS_R15 * WORD_SIZE
34 .section .noinstr.text, "ax"
37 * __vmx_vcpu_run - Run a vCPU via a transition to VMX guest mode
38 * @vmx: struct vcpu_vmx *
39 * @regs: unsigned long * (to guest registers)
40 * @flags: VMX_RUN_VMRESUME: use VMRESUME instead of VMLAUNCH
41 * VMX_RUN_SAVE_SPEC_CTRL: save guest SPEC_CTRL into vmx->spec_ctrl
44 * 0 on VM-Exit, 1 on VM-Fail
46 SYM_FUNC_START(__vmx_vcpu_run)
48 mov %_ASM_SP, %_ASM_BP
60 /* Save @vmx for SPEC_CTRL handling */
63 /* Save @flags for SPEC_CTRL handling */
67 * Save @regs, _ASM_ARG2 may be modified by vmx_update_host_rsp() and
68 * @regs is needed after VM-Exit to save the guest's register values.
72 /* Copy @flags to BL, _ASM_ARG3 is volatile. */
75 lea (%_ASM_SP), %_ASM_ARG2
76 call vmx_update_host_rsp
78 ALTERNATIVE "jmp .Lspec_ctrl_done", "", X86_FEATURE_MSR_SPEC_CTRL
81 * SPEC_CTRL handling: if the guest's SPEC_CTRL value differs from the
82 * host's, write the MSR.
84 * IMPORTANT: To avoid RSB underflow attacks and any other nastiness,
85 * there must not be any returns or indirect branches between this code
88 mov 2*WORD_SIZE(%_ASM_SP), %_ASM_DI
89 movl VMX_spec_ctrl(%_ASM_DI), %edi
90 movl PER_CPU_VAR(x86_spec_ctrl_current), %esi
93 mov $MSR_IA32_SPEC_CTRL, %ecx
101 * Since vmentry is serializing on affected CPUs, there's no need for
102 * an LFENCE to stop speculation from skipping the wrmsr.
105 /* Load @regs to RAX. */
106 mov (%_ASM_SP), %_ASM_AX
108 /* Check if vmlaunch or vmresume is needed */
109 testb $VMX_RUN_VMRESUME, %bl
111 /* Load guest registers. Don't clobber flags. */
112 mov VCPU_RCX(%_ASM_AX), %_ASM_CX
113 mov VCPU_RDX(%_ASM_AX), %_ASM_DX
114 mov VCPU_RBX(%_ASM_AX), %_ASM_BX
115 mov VCPU_RBP(%_ASM_AX), %_ASM_BP
116 mov VCPU_RSI(%_ASM_AX), %_ASM_SI
117 mov VCPU_RDI(%_ASM_AX), %_ASM_DI
119 mov VCPU_R8 (%_ASM_AX), %r8
120 mov VCPU_R9 (%_ASM_AX), %r9
121 mov VCPU_R10(%_ASM_AX), %r10
122 mov VCPU_R11(%_ASM_AX), %r11
123 mov VCPU_R12(%_ASM_AX), %r12
124 mov VCPU_R13(%_ASM_AX), %r13
125 mov VCPU_R14(%_ASM_AX), %r14
126 mov VCPU_R15(%_ASM_AX), %r15
128 /* Load guest RAX. This kills the @regs pointer! */
129 mov VCPU_RAX(%_ASM_AX), %_ASM_AX
131 /* Check EFLAGS.ZF from 'testb' above */
135 * After a successful VMRESUME/VMLAUNCH, control flow "magically"
136 * resumes below at 'vmx_vmexit' due to the VMCS HOST_RIP setting.
137 * So this isn't a typical function and objtool needs to be told to
138 * save the unwind state here and restore it below.
143 * If VMRESUME/VMLAUNCH and corresponding vmexit succeed, execution resumes at
144 * the 'vmx_vmexit' label below.
154 _ASM_EXTABLE(.Lvmresume, .Lfixup)
155 _ASM_EXTABLE(.Lvmlaunch, .Lfixup)
157 SYM_INNER_LABEL(vmx_vmexit, SYM_L_GLOBAL)
159 /* Restore unwind state from before the VMRESUME/VMLAUNCH. */
163 /* Temporarily save guest's RAX. */
166 /* Reload @regs to RAX. */
167 mov WORD_SIZE(%_ASM_SP), %_ASM_AX
169 /* Save all guest registers, including RAX from the stack */
170 pop VCPU_RAX(%_ASM_AX)
171 mov %_ASM_CX, VCPU_RCX(%_ASM_AX)
172 mov %_ASM_DX, VCPU_RDX(%_ASM_AX)
173 mov %_ASM_BX, VCPU_RBX(%_ASM_AX)
174 mov %_ASM_BP, VCPU_RBP(%_ASM_AX)
175 mov %_ASM_SI, VCPU_RSI(%_ASM_AX)
176 mov %_ASM_DI, VCPU_RDI(%_ASM_AX)
178 mov %r8, VCPU_R8 (%_ASM_AX)
179 mov %r9, VCPU_R9 (%_ASM_AX)
180 mov %r10, VCPU_R10(%_ASM_AX)
181 mov %r11, VCPU_R11(%_ASM_AX)
182 mov %r12, VCPU_R12(%_ASM_AX)
183 mov %r13, VCPU_R13(%_ASM_AX)
184 mov %r14, VCPU_R14(%_ASM_AX)
185 mov %r15, VCPU_R15(%_ASM_AX)
188 /* Clear return value to indicate VM-Exit (as opposed to VM-Fail). */
192 /* Discard @regs. The register is irrelevant, it just can't be RBX. */
196 * Clear all general purpose registers except RSP and RBX to prevent
197 * speculative use of the guest's values, even those that are reloaded
198 * via the stack. In theory, an L1 cache miss when restoring registers
199 * could lead to speculative execution with the guest's values.
200 * Zeroing XORs are dirt cheap, i.e. the extra paranoia is essentially
201 * free. RSP and RBX are exempt as RSP is restored by hardware during
202 * VM-Exit and RBX is explicitly loaded with 0 or 1 to hold the return
223 * IMPORTANT: RSB filling and SPEC_CTRL handling must be done before
224 * the first unbalanced RET after vmexit!
226 * For retpoline or IBRS, RSB filling is needed to prevent poisoned RSB
227 * entries and (in some cases) RSB underflow.
229 * eIBRS has its own protection against poisoned RSB, so it doesn't
230 * need the RSB filling sequence. But it does need to be enabled, and a
231 * single call to retire, before the first unbalanced RET.
234 FILL_RETURN_BUFFER %_ASM_CX, RSB_CLEAR_LOOPS, X86_FEATURE_RSB_VMEXIT,\
235 X86_FEATURE_RSB_VMEXIT_LITE
237 pop %_ASM_ARG2 /* @flags */
238 pop %_ASM_ARG1 /* @vmx */
240 call vmx_spec_ctrl_restore_host
242 /* Put return value in AX */
243 mov %_ASM_BX, %_ASM_AX
259 cmpb $0, kvm_rebooting
263 /* VM-Fail: set return value to 1 */
267 SYM_FUNC_END(__vmx_vcpu_run)
273 * vmread_error_trampoline - Trampoline from inline asm to vmread_error()
274 * @field: VMCS field encoding that failed
275 * @fault: %true if the VMREAD faulted, %false if it failed
277 * Save and restore volatile registers across a call to vmread_error(). Note,
278 * all parameters are passed on the stack.
280 SYM_FUNC_START(vmread_error_trampoline)
282 mov %_ASM_SP, %_ASM_BP
296 /* Load @field and @fault to arg1 and arg2 respectively. */
297 mov 3*WORD_SIZE(%_ASM_BP), %_ASM_ARG2
298 mov 2*WORD_SIZE(%_ASM_BP), %_ASM_ARG1
302 /* Zero out @fault, which will be popped into the result register. */
303 _ASM_MOV $0, 3*WORD_SIZE(%_ASM_BP)
319 SYM_FUNC_END(vmread_error_trampoline)
321 SYM_FUNC_START(vmx_do_interrupt_nmi_irqoff)
323 * Unconditionally create a stack frame, getting the correct RSP on the
324 * stack (for x86-64) would take two instructions anyways, and RBP can
325 * be used to restore RSP to make objtool happy (see below).
328 mov %_ASM_SP, %_ASM_BP
332 * Align RSP to a 16-byte boundary (to emulate CPU behavior) before
333 * creating the synthetic interrupt stack frame for the IRQ/NMI.
341 CALL_NOSPEC _ASM_ARG1
344 * "Restore" RSP from RBP, even though IRET has already unwound RSP to
345 * the correct value. objtool doesn't know the callee will IRET and,
346 * without the explicit restore, thinks the stack is getting walloped.
347 * Using an unwind hint is problematic due to x86-64's dynamic alignment.
349 mov %_ASM_BP, %_ASM_SP
352 SYM_FUNC_END(vmx_do_interrupt_nmi_irqoff)